Pre-Trained Gas Sensing for OGE and Thermal Runaway Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to accurately distinguish between off-gas events (OGE) and thermal runaway events (TRE) from non-OGE interfering gases without requiring continuous training or reference sensors, posing safety risks in battery systems.
Innovation Solution
Pre-training gas sensors using machine learning (ML) or deep learning (DL) algorithms to generate unique sensor signals, extract features, and establish decision boundaries for OGE and TRE, eliminating the need for field training or reference sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gas sensing systems are used without pre-training, then the device complexity is reduced, but the measurement precision and reliability deteriorate due to inability to distinguish OGE/TRE from interfering gases
Solution Approach 1:
The gas sensor is pre-trained a priori using machine learning or deep learning algorithms before field deployment to classify different gas types. This preliminary training enables the sensor to distinguish OGE/TRE from interfering gases without requiring continuous training or complex reference sensor systems during operation.
Solution Approach 2:
The pre-trained gas sensor autonomously classifies gas analytes by itself without requiring external reference sensors or continuous human intervention. The sensor uses its own sensor signals and embedded decision boundaries to independently distinguish between different gas types, making the system self-sufficient.
2Reliability
If reference sensors or continuous training systems are implemented, then the measurement precision improves, but the device complexity and loss of time increase
Solution Approach 1:
All training is completed a priori before field deployment. The sensor learns to classify gases during manufacturing or initialization, eliminating the need for continuous training during operation. This preliminary action ensures high reliability without time loss during actual gas monitoring.
Solution Approach 2:
The training process is extracted from the operational phase and moved to the pre-deployment phase. By separating training from operation, the system achieves high reliability during gas monitoring without requiring time-consuming training during field use.
3Measurement precision
If multiple reference sensors are used to distinguish gas types, then the measurement precision improves, but the device complexity and manufacturing cost increase
Solution Approach 1:
A single gas sensor performs gas type discrimination by itself through pre-trained machine learning algorithms. The sensor processes its own sensor signals and uses embedded decision boundaries to classify gases, eliminating the need for multiple reference sensors or complex sensor arrays.
Solution Approach 2:
The physical system of multiple reference sensors is replaced with an information-processing system. Instead of using multiple sensors to physically detect different gases, a single sensor uses machine learning algorithms to computationally distinguish gas types based on sensor signal patterns.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables real-time detection and classification of OGE and TRE without additional training, reducing false positives and enhancing safety by providing early warnings for thermal runaway conditions.
Implementation Method 1
monitoring the gas source for release of a gas analyte, by at least one gas sensor having one or more sensing electrodes
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
This disclosure relates to systems and methods for monitoring and classifying released gases in an enclosed system having a gas source, by a gas sensor that has been a priori pre-trained to distinguish an off-gas event (OGE) or a thermal run off event (TRE) from non-OGE interfering gases release. The pre-training utilizes one of a machine learning (ML) or a deep learning (DL) algorithm to pre-train the gas sensor to detect a plurality of known gas analyte to generate sensor signals with respective unique characteristics, extracting features from the sensor signals to establish a decision boundary or an estimated probability of a false positive release of the OGE or the TRE from the non-OGE type of interfering gas release. The established decision boundaries or probability distributions are implemented as candidate model for field deployment to classify the released gases to distinguish whether OGE or TRE takes place.